What happened
Google has documented a personal-avatar feature in Gemini Apps. A user can record their face and voice, creating an avatar that can be used in Gemini image generation with Nano Banana and video generation with Gemini Omni. Google describes the goal as streamlining content creation while allowing friends, family, and collaborators to include the creator’s likeness in their own generated images and videos.
Creation begins on a computer at Gemini, where the user selects the Avatar option from the file-upload controls. The user then scans a QR code with a phone or tablet and follows the recording flow there, granting camera and microphone access to record face and voice. They return to the computer to finish and select “Use Avatar.” Google specifies practical recording conditions: the phone should be held at eye level; lighting should not be too dark or bright; the eyes, nose, and mouth must be visible; the recording should happen in a quiet environment; and the background should not contain people or images of faces. The account owner must complete the process, and Google says minors must not record themselves.
In a Gemini prompt, an available avatar appears as an @ reference using the relevant Google username. Google’s examples include generating an image of the person singing with an orchestra, writing a story in which they are the main character, or showing the prompt author and another named avatar dancing. A personal avatar can be used for image and text-related prompts under the described Gemini flow. Creating video with a personal avatar, however, requires a Google AI plan.
The new sharing flow lets an avatar owner enter another person’s Gmail address from Gemini’s Avatar settings. The recipient can accept or decline the shared avatar; acceptance is required before the recipient can use that likeness in images or videos. Owners can inspect the list of recipients, remove a recipient, and update or remove an expiration date for a recipient’s access. Revocation prevents that person from making new images or videos using the avatar, but does not erase content they already made with the owner’s likeness or voice. Deleting the avatar removes the recorded selfie and voice data from Google’s systems and stops new content creation with that avatar, while already created or published content remains.
The feature has clear access constraints. Creating an avatar requires a personal Google Account and a user who is at least 18. Avatars are currently unavailable in the EEA, Switzerland, and the United Kingdom, and are currently supported only in English.
What the archive adds
This is a new fact on a known Gemini generative-media story. Earlier Field Notes covered Gemini 3.8 Live with Live Avatar as a continuous audiovisual agent interface and Gemini 4 Argon as a broader availability move across Google surfaces. The present change is different: it makes a user’s recorded identity shareable and administrable by other users.
Trend signal
This modestly supports Generative media becomes application infrastructure. The useful capability is no longer only asking a model for a one-off image: a likeness becomes a reusable input that can be selected, shared, accepted, time-limited, and revoked inside a product workflow.
Why it matters
Teams building AI media tools often treat identity consistency as a quality problem: preserve a face, voice, or character across outputs. Google’s design shows that it is also an authorization problem. The system makes a reusable identity asset, then surrounds it with named recipients, an explicit acceptance step, expiry settings, visibility into access, and a revocation control.
That does not solve every provenance or safety question. In particular, revoking access does not remove prior outputs. But it gives product teams a concrete minimum distinction to preserve: permission to create new media is separable from the persistence of media already created. Any system that lets collaborators generate content with a person’s recognizable identity will need to decide where that separation appears in its own interface, logs, storage, and policy.
Limits and uncertainty
Google’s help page describes the feature and its controls, not its underlying identity-verification method, abuse detection, watermarking, audit trail, or enforcement process. It does not establish how quickly removal or expiry propagates across all Gemini surfaces, whether a recipient can export material before access is revoked, or how content made with a shared avatar is labelled. The page also does not give pricing for the required Google AI plan, rollout dates, supported regions beyond the listed exclusions, or evidence that the controls reduce misuse in practice.
Skill to build now
Map the consent lifecycle for one identity-bearing asset in your product this week. Define who can grant access, what the recipient must explicitly accept, where an owner can see active grantees, whether access can expire, and exactly what revocation stops. Then test the uncomfortable case Google makes explicit: a user revokes access after collaborators have already generated and exported media. Your product should state clearly what remains available and what cannot be undone.
Reader takeaway
Gemini is turning a recorded face and voice into a shareable generative-media input, but it pairs that input with explicit recipient acceptance, expiry, and revocation. For builders, the notable lesson is operational: reusable identity requires lifecycle controls, not just better image or video consistency.
Open question
Will Google extend these controls with inspectable provenance and stronger handling of media produced before an owner revokes access?